Automate a daily CI failure digest from GitHub Actions

One failed run is a signal. Fifteen in a week from the same step are noise that people learn to ignore, and the real break hides inside. The useful view is not each failure but the pattern: which workflow, which step, which branch, and how often it repeats. That view is what a developer builds by scanning the Actions tab, and it is exactly the chore that stops when the team is busy. A daily digest keeps the same view available on the days nobody has time to look.

You add the job and connect the repository with read access. Each morning a scheduled agent lists the workflow runs for the repository, filtered to a created date range covering yesterday and to the failure status, up to one hundred runs per page. For every failed run it lists the jobs, reads the conclusion of each step to find where it broke, and groups runs by workflow, failing step and branch. The Branches field keeps every branch in the digest or narrows it to the default branch, which leaves pull request noise out. It marks a failure as a repeat when the same workflow and step fail in several runs, and it fetches a log when a recurring pattern needs explaining. It does not try to decide whether a failure is flaky or broken; it reports the pattern and leaves that call to you.

The CTO agent, as department head, reviews the digest before it goes out: do the counts add up, is a failure on a feature branch kept apart from one on the default branch, is a suggested action safe to take. Then you approve, edit or skip it, and nothing is posted or re-run until you do. On a day with no failed runs, the digest says so in one line, so a quiet morning is confirmed rather than assumed. Start with one repository, and read the digest for a week against your own scan. Once the digest matches how you would have read the Actions tab, add the others.